Twitch Turns On Amazon AI Training by Default: ‘Nobody Would Opt In’

Twitch has started allowing Amazon to train generative AI models on the work of the people who stream on it, with the setting enabled by default.

The Amazon-owned platform announced the change on Wednesday as the addition of a control, telling users it had added a setting that lets them opt out of having channel content used to train generative AI models across Amazon. The framing carried the news that the training was happening at all, though streamers had reported in late July that “multiple sources” had told them the move was imminent.

Everything on a channel is in scope, with streams, videos on demand, clips, highlights, chat, text and images all usable to train models whose purpose, in Twitch’s words, is to generate or synthesize text, audio, images or video. Its own FAQ offers the example of a streamer’s audio being used to refine speech-to-text models, improving captions on Twitch and across Amazon. 

The toggle, labeled “Training for Generative AI,” sits under Security and Privacy at the bottom of the account settings page.

‘Nobody would opt in’

Head of community Mary Kish and chief product officer Mike Minton took questions on the company’s own channel on Wednesday, to an audience of nearly 3,000 and a chat filling with anti-AI messages. Asked why the setting was not opt-in, Minton answered plainly: “If it was opt-in, nobody would opt in. That’s honestly the answer.”

Asked by a user whether their videos had already been used for training, Minton said he did not actually know, because he did not know what Amazon had done in terms of model training or what it had and had not used. Kish said she had personally opted out, and described the control itself as a response to a community that does not want its work training generative models. Minton said he was comfortable with the training and would not opt out were he a streamer.

Pressed on what happens to someone who has opted out but appears in the stream of someone who has not, Minton had no answer beyond calling it a “good question,” according to Kotaku, though the FAQ published alongside the change states that the streamer’s preference governs whether a viewer’s chat can be used. He confirmed there are no plans to let users see which channels are opted in.

Opting out is also narrower than it sounds, since Twitch can still run AutoMod and generate captions, which it says do not retain content to train models that produce new material. Some of those functions protect the whole platform, the company says, so turning them off for one person is not offered.

The change lands as AI companies run short of training material and turn to the platforms they already own. TechCrunch noted that thousands of hours of a streamer’s recorded voice and face make Twitch an unusually rich source, and that many creators may hand it over without realizing.